Unaligned AI

Expert assessments identify advanced artificial intelligence as the highest-probability route to human extinction or permanent civilizational collapse this century. Toby Ord estimates roughly a 10% chance of existential catastrophe from unaligned AI over 100 years, exceeding engineered pandemics (~3%) and far above nuclear war or extreme climate change (~0.1% each). AI researcher surveys commonly show median probabilities of 5% or higher for extremely bad outcomes including extinction, with many specialists at 10–20% or more. A 2026 Lancet Commission singles out malicious AI use (such as enabling bioweapons) and nuclear conflict as the only threats with realistic extinction potential by 2100, with specialist estimates around 3% for AI-driven extinction. Rapid capability advances, competitive racing among major powers, and lagging alignment and governance increase the chance that powerful systems pursue goals incompatible with human survival through loss of control, takeover, or amplification of other risks.